Before calling CNVs, we performed analyses on the variability in intensities across the probes for each array. This way, we could identify outlier experiments for each platform and also calculate summary measures of variability for the different arrays before CNV calling. We inspected platform-specific quality control measures including (i) mean, median and s.d. of log R ratio and B allele frequency for Illumina arrays, (ii) contrast quality control and median absolute pair-wise differences for Affymetrix arrays and (iii) producer-derived derivative log2 ratio spread for CGH arrays. In addition, measures of variability were calculated on the raw data for all platforms (Supplementary Methods), including the derivative log2 ratio spread and interquartile range (Table 2). The derivative log2 ratio spread statistic describes the absolute value of the log2 ratio variance from each probe to the next, averaged over the entire genome. The interquartile range is a measure of the dispersion of intensities in the center of the distribution, and is therefore less sensitive to outliers. The variability estimates include both biological and technical variability, but the effect of biological variability should be small on global statistics.
